Project Summary The large variability in lesions, impairment, and responsiveness to training following stroke has hindered the development of principled and cost-effective approaches to neuro-rehabilitation of the upper extremity. Our long-term goal is to develop predictive personalized neurorehabilitation therapy based on large data sets. This proposal is based on a unique opportunity to design and execute a large neuro-rehabilitation cohort study at a relatively low cost. Building on our established US-French collaborations, with interdisciplinary expertise in neurorehabilitation, brain imaging, dynamical systems, and statistical learning, we will predict recovery and individualize therapy with the following novel three-pronged approach. In Aim 1, we will develop a database of clinical and neural patient characteristics, treatments, and outcomes from 500 patients post-stroke receiving upper extremity rehabilitation therapy with the ARMEO Spring device (a gravity compensating exoskeleton) in routine clinical care. Inclusion criteria will be as broad as possible to include patients with a large variety of brain lesions, as assessed by state- of-the-art magnetic resonance imaging (MRI) and functional MRI scans. In aim 2, using the database, we will predict long-term changes in upper extremity outcomes as a function of patient's characteristics and treatment using dynamical models that link motor learning to recovery. The final models will expand and combine previous computational models of motor learning at small time scales with models of recovery at long time scales, and will include mixed effects to accurately predict long-term recovery for individual patients. In aim 3, based on these predictions, we will perform a feasibility study aimed at individualizing upper extremity rehabilitation to maximize recovery. Given a new patient, characterized by a number of baseline characteristics that predict recovery, we will select the schedule of treatment that was the most effective for similar patients in the database. The recovery models and scheduling methods developed in this proposal will provide the basis for future clinical software that suggests timing, dosage, and content of therapy from early clinical data, kinematic performance, and routine scans. Such an approach will transform neurorehabilitation programs because the clinician, patient, and insurance company will be able to determine effective treatments while reducing costs.